In [1]:
import os, sys
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import torch
from torch import nn
from torch.utils.data import DataLoader, Dataset
import torchvision.transforms as transforms
import torchvision.models as models
from torchvision import transforms, utils
from torch import optim
from torch import cuda
from torchvision import transforms
from sklearn.model_selection import StratifiedKFold
from PIL import Image
import cv2
import random
import math
import gc, pickle
from tqdm import tqdm_notebook as tqdm
import warnings
warnings.filterwarnings('ignore')
In [2]:
base_path = "../input/anime-faces/data/"
img_path = os.listdir(base_path)
img_path = [path for path in img_path if 'png' in path]
In [3]:
def read_img(path):
base_path = "../input/anime-faces/data/"
img = cv2.imread(f"{base_path}{path}")
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
return img
In [4]:
len(img_path)
Out[4]:
In [5]:
plt.figure(figsize=(30,10))
sample_path = random.sample(img_path, 15)
for i, path in enumerate(sample_path):
plt.subplot(3,5,i+1)
img = read_img(path)
img = cv2.normalize(img, None, alpha=0, beta=1.0, norm_type=cv2.NORM_MINMAX, dtype=cv2.CV_32F)
plt.imshow(img)
plt.show()
In [6]:
sorted([
f"../input/dcgan-generate-anime-faces-4/{a}" for a in\
os.listdir('../input/dcgan-generate-anime-faces-4/') if "gene" in a])
Out[6]:
In [7]:
class Generator(nn.Module):
def __init__(self, img_size, latent_dim, channels):
super(Generator, self).__init__()
self.init_size = img_size // 8
self.conv_init_dim = 512
self.l1 = nn.Sequential(nn.Linear(latent_dim, self.conv_init_dim * self.init_size ** 2))
self.channels = channels
self.conv_blocks = nn.Sequential(
nn.BatchNorm2d(self.conv_init_dim),
nn.Upsample(scale_factor=2),
nn.Conv2d(self.conv_init_dim, self.conv_init_dim, 3, stride=1, padding=1),
nn.BatchNorm2d(self.conv_init_dim, 0.8),
nn.LeakyReLU(0.2, inplace=True),
nn.Upsample(scale_factor=2),
nn.Conv2d(self.conv_init_dim, self.conv_init_dim//2, 3, stride=1, padding=1),
nn.BatchNorm2d(self.conv_init_dim//2, 0.8),
nn.LeakyReLU(0.2, inplace=True),
nn.Conv2d(self.conv_init_dim//2, self.conv_init_dim//2, 3, stride=1, padding=1),
nn.BatchNorm2d(self.conv_init_dim//2, 0.8),
nn.LeakyReLU(0.2, inplace=True),
nn.Upsample(scale_factor=2),
nn.Conv2d(self.conv_init_dim//2, self.conv_init_dim//2, 3, stride=1, padding=1),
nn.BatchNorm2d(self.conv_init_dim//2, 0.8),
nn.LeakyReLU(0.2, inplace=True),
nn.Conv2d(self.conv_init_dim//2, self.conv_init_dim//4, 3, stride=1, padding=1),
nn.BatchNorm2d(self.conv_init_dim//4, 0.8),
nn.LeakyReLU(0.2, inplace=True),
nn.Conv2d(self.conv_init_dim//4, self.channels, 3, stride=1, padding=1),
nn.Sigmoid(),
)
def forward(self, z):
out = self.l1(z)
out = out.view(out.shape[0], self.conv_init_dim, self.init_size, self.init_size)
img = self.conv_blocks(out)
return img
img_size = 64
latent_dim = 128
channels=3
b_size=40
paths = ['../input/dcgan-generate-anime-faces-4/generator337.pt',
'../input/dcgan-generate-anime-faces-4/generator1348.pt',
'../input/dcgan-generate-anime-faces-4/generator2359.pt',
'../input/dcgan-generate-anime-faces-4/generator3370.pt',
'../input/dcgan-generate-anime-faces-4/generator4381.pt',
'../input/dcgan-generate-anime-faces-4/generator5392.pt',
'../input/dcgan-generate-anime-faces-4/generator6403.pt',
'../input/dcgan-generate-anime-faces-4/generator7414.pt',
'../input/dcgan-generate-anime-faces-4/generator8425.pt',
'../input/dcgan-generate-anime-faces-4/generator9436.pt',
'../input/dcgan-generate-anime-faces-4/generator10110.pt',]
generator = Generator(img_size, latent_dim, channels)
z = np.random.normal(0, 1, (b_size, latent_dim)).tolist()
z = torch.FloatTensor(z)
for path in paths:
state_dict = torch.load(path)
generator.load_state_dict(state_dict)
imgs = generator(z)
print(imgs.size())
imgs = imgs.detach()
plt.figure(figsize=(30,14))
grid = utils.make_grid(imgs, nrow=10)
plt.imshow(grid.numpy().transpose((1, 2, 0)))
plt.title(path.split('/')[-1])
plt.show()
